RESEARCH ARTICLE

Muscle cocontraction following dynamics learning

Mohammad Darainy Æ David J. Ostry

Received: 17 March 2008 / Accepted: 2 June 2008 / Published online: 27 June 2008

� Springer-Verlag 2008

Abstract Coactivation of antagonist muscles is readily

observed early in motor learning, in interactions with

unstable mechanical environments and in motor system

pathologies. Here we present evidence that the nervous

system uses coactivation control far more extensively and

that patterns of cocontraction during movement are closely

tied to the specific requirements of the task. We have

examined the changes in cocontraction that follow

dynamics learning in tasks that are thought to involve

finely sculpted feedforward adjustments to motor com-

mands. We find that, even following substantial training,

cocontraction varies in a systematic way that depends on

both movement direction and the strength of the external

load. The proportion of total activity that is due to co-

contraction nevertheless remains remarkably constant.

Moreover, long after indices of motor learning and elec-

tromyographic measures have reached asymptotic levels,

cocontraction still accounts for a significant proportion of

total muscle activity in all phases of movement and in all

load conditions. These results show that even following

dynamics learning in predictable and stable environments,

cocontraction forms a central part of the means by which

the nervous system regulates movement.

Keywords Motor learning � EMG � Impedance control

Introduction

Muscles in the human motor system have the unique

characteristic, that as force generating devices, they

actively work only in a single direction. Hence, for each

single mechanical degree-of-freedom, at least two antago-

nistic muscles are necessary to control motion. The ways in

which these antagonistic muscles are recruited in different

tasks is a fundamental problem in understanding human

motor control. Two distinct control mechanisms have been

identified in motor control studies. In one case, the nervous

system reciprocally activates sets of antagonist muscles

and produces torques at a desired joint. Alternatively,

antagonistic muscles are recruited at the same time. When

cocontraction is balanced there is no kinematic effect, but

there are resulting changes in the mechanical impedance of

the musculoskeletal system. Reciprocal control and muscle

cocontraction have been studied extensively. However, the

relative contribution of each to different motor tasks is not

well understood nor is manner in which cocontraction

varies with the specific requirements of the task.

It has been shown that subjects use cocontraction control

to offset the effects of destabilizing forces both under static

conditions (Darainy et al. 2004) and during reaching

movements (Burdet et al. 2001; Franklin et al. 2003a).

Cocontraction control has been documented in the early

stages of motor learning. It was shown that the cocon-

traction of muscles declines as learning progresses, both

when subjects learn stable dynamics (Franklin et al. 2003b;

Thoroughman and Shadmehr 1999) and for unstable

dynamical tasks (Milner and Cloutier 1993). Little is

known about the characteristics of cocontraction following

M. Darainy � D. J. Ostry (&)

Department of Psychology, McGill University, 1205 Dr. Penfield

Avenue, Montreal, QC, Canada H3A 1B1

e-mail: ostry@motion.psych.mcgill.ca

M. Darainy

Shahed University, Tehran, Iran

D. J. Ostry

Haskins Laboratories, New Haven, CT, USA

123

Exp Brain Res (2008) 190:153–163

DOI 10.1007/s00221-008-1457-y

motor learning in stable environments. Apart from a gen-

eral decline in cocontraction following learning, the

manner in which cocontraction during movement is tuned

to specific requirements of the task is unknown. In the

present study, we have assessed the role of muscle

cocontraction following adaptation to stable environmental

dynamics in two situations. We have first examined the

degree to which cocontraction during movement is modu-

lated with the strength of the external force field. The

modulation of cocontraction with the level of external force

would be consistent with the idea that cocontraction control

is part of the way in which the nervous system compensates

for the effects of external dynamics. A second focus is

whether cocontraction is differentially tuned to offset the

effects of loads associated with movements in different

directions. Different muscle pairs are involved in moving

in different directions and accordingly different patterns of

muscle coactivation may be necessary to optimally support

these movements. Apart from demonstrations under static

conditions (Gribble and Ostry 1998; Gomi and Osu 1998),

it is unknown whether the nervous system is capable of

modulating the balance of cocontraction over different

muscle pairs to counteract the forces involved in different

movement directions.

In the present study we have assessed the role of

muscle cocontraction after extensive practice. We have

used a center-out reaching task with a clockwise velocity-

dependent curl field to gauge this effect (Shadmehr and

Mussa-Ivaldi 1994). Two primary directions of movement

were chosen for this study. Reaching movements in one

direction required mostly elbow rotation (elbow move-

ment condition) while movements in the second direction

involved primarily shoulder rotation (shoulder movement

condition). In each direction, subjects trained with two

levels of force-field strength consecutively. We obtained a

measure of muscle cocontraction using a technique

described previously—the minimum normalized muscle

activity for each antagonist pair of muscles (Thoroughman

and Shadmehr 1999; Gribble et al. 2003). This is a

measure of co-occurring activity that is shared by an

antagonist muscle pair and provides an estimate of

cocontraction. We observed that even following adapta-

tion and after considerable training, the cocontraction

level varied with the strength of the force-field. We also

observed that movement direction had a substantial

influence on the pattern of cocontraction. Moreover, the

cocontraction associated with these effects accounted for a

substantial and remarkably constant portion of total

muscle activation. These results show that even during

wholly stable interactions with the environment, cocon-

traction is modulated with the level of the force-field and

also with the direction of movement. These findings

suggest that coactivation control is an integral component

of the motor control system and is applied broadly even

after learning.

Methods

Subjects and apparatus

Ten male subjects, between 19 and 28, participated in this

study. All subjects were right-handed and had no history of

neuromuscular disorder. Experimental procedures were

approved by the McGill University Ethics Committee.

Subjects were seated in front of a two degree-of-freedom

planar robotic arm (Inmotion2, Interactive Motion Tech-

nologies Inc.) and held the handle in their right hand. For

each subject, the height of the seat was adjusted to produce

an abduction angle at the shoulder of 85�. The seat position

was also adjusted to have a shoulder angle of 45�, relative

to the frontal plane, and an elbow angle of 90�, relative to

the upper arm, as the start point for all movements. To

decrease friction, the subject’s hand was supported on the

surface of a glass table by an air-sled. Hand position was

measured with optical encoders at the robot joints.

Experimental task

The experimental session was divided into a familiarization

phase and an experimental phase. The electrodes were

placed just prior to the experimental phase. In the famil-

iarization phase of experiment, subjects were trained to

make 15 cm reaching movements in 600 ± 50 ms to two

visual targets that were positioned just below the surface of

the glass table. No forces were applied during this part of

the experiment (null condition). One visual target was

placed lateral to the subject at the left of the start point. The

reaching movement to this target involved shoulder flexion

(on average 21�) and to a lesser extent elbow extension

(6�). We will refer to this as the shoulder movement con-

dition. A second target was placed in a diagonal direction,

forward and to the right of the start point. Movement to this

target primarily involved elbow extension (30�, combined

with 6� shoulder flexion) and will be referred to as the

elbow movement condition.

Subjects were asked to move as straight as possible and

audio–visual feedback of movement duration was provided

at the end of each trial. In familiarization phase, two blocks

of 50 trials each were carried out to each of the targets.

Subjects had full view of the arm at all times during the

experiment. Each trial started with a short beep and sub-

jects were asked to reach the target in a single movement.

They were also told that their reaction time was not the

focus of the study and accordingly was not included in the

measured movement duration. Following movement, when

154 Exp Brain Res (2008) 190:153–163

123

the hand stopped inside a 1 cm diameter circular target

zone, visual feedback of their movement duration was

displayed on a monitor beside the robot. They also heard

audio feedback at the same time. The robotic arm then

moved the subject’s hand back to the start point and after a

500 ms delay the next trial began.

Six shoulder and elbow muscle sites were identified and

prepared for electrode placement. The muscles of interest

were two single-joint shoulder muscles (pectoralis clavic-

ular head and posterior deltoid), two muscles acting at the

elbow (the double-joint muscle biceps long head and single

joint triceps lateral head) and two bi-articular shoulder and

elbow muscles (biceps short head and triceps long head). A

variety of test maneuvers were carried out to determine

electrode placement. EMG signals were amplified and

analog filtered between 20 and 450 Hz (Delsys, Bangoli 8).

In the experimental phase of study, subjects were ran-

domly divided into two groups of five subjects each. The

first group was tested first with movements to the lateral

target (shoulder movement condition) and then repeated

the same procedure with movements to the diagonal target

(elbow movement condition). The second group of subjects

did the same task but in opposite order. Five blocks of

reaching movements were carried out to each target.

Blocks 1 and 2 involved 50 trials each and were carried out

under null conditions (only Block 2 was recorded). Block 3

involved 150 trials under low force-field conditions (see

below). A fourth block (Block 4) also involved 150 trials

under high force-field conditions. A final block of 50 trials

was carried out under null conditions to test for afteref-

fects. EMG data and kinematics were recorded

simultaneously.

Clock-wise velocity-dependent forces were used during

force-field trials. Equation 1 shows the forces that were

applied to the subject’s hand.

fx

fy

� �¼ b

0 1

�1 0

� �vx

vy

� �ð1Þ

In this equation, fx and fy are the commanded force to the

robot, vx and vy are hand velocities in Cartesian coordinates

and b gives the field strength. b was set at 7 N s/m for the

low force-field condition and 17 N s/m for the high force-

field condition. In 20% of randomly chosen force-field

trials, the robot motors were turned-off and subjects were

tested under null conditions (catch trials).

Data analysis

Position signals were sampled at 200 Hz, low-pass

Butterworth filtered at 20 Hz and numerically differenti-

ated to produce the velocity signals. EMG signals were

sampled at 1,000 Hz and digitally band-bass filtered

between 20 and 450 Hz. EMG signals were then full-wave

rectified and the root-mean-square of the signal was cal-

culated over a sliding window of 100 ms. A measure of

baseline EMG activity was removed from all signals on a

per muscle basis. The measure was obtained by recording a

single 30-s trial at the beginning of the experimental phase,

during which the subject rested at the center of workspace

and did nothing. For each trial, the EMG signals and

kinematics data were time-aligned at movement start.

Movement onset was scored at the time that tangential

hand velocity exceeded 20 mm/s. Movement end was also

scored when the hand tangential velocity fell below the 20

mm/s. Kinematic error was used to assess learning and was

defined on a trial-by-trial basis, as the maximum perpen-

dicular distance (PD) between the hand trajectory and a

straight line that connected movement start and end.

We also assessed performance by computing a learning

index (Hwang et al. 2003), LI, a measure that takes into

account both force-field trials and catch-trials and, hence,

corrects for possible differences in performance due to

differences in the action of the force field. The LI is defined

as:

LI ¼ PDcatchj jPDcatchj j þ PDffj j ð2Þ

where PDff is the maximum perpendicular distance from a

straight line under force field conditions and PDcatch is the

same measure for catch trials. The learning index ranges

from 0.0 early in training to an upper limit of 1.0 under

conditions of complete adaptation.

Following the removal of baseline activity, measures of

EMG were normalized on a muscle-by-muscle basis for

each subject separately. The normalization serves to equate

EMG magnitudes of antagonist muscles under static con-

ditions. As a normalization factor we used the mean muscle

activity from the high force condition during a 200 ms

window from 500 to 300 ms before movement onset

(averaged over the 300 trials in both movement directions).

As a control, we repeated the normalization procedure

using the mean EMG activity in a 200 ms window starting

300 ms after the end of movement. We also repeated the

normalization procedure based on measures of maximum

voluntary cocontraction that were recorded prior to the

experimental measurements. In both cases, the results

obtained were qualitatively similar to those reported above.

We have not used the more usual technique of nor-

malizing EMG relative to maximum voluntary contraction

(MVC). In order to obtain a normalized measure of muscle

cocontraction (as opposed to individual muscle contrac-

tion), we needed a reference that equated for EMG levels in

antagonist muscles when the arm is cocontracted in statics.

Conventional MVC measures are obtained for each muscle

separately and MVC values for antagonist muscle pairs are

unlikely to result in static equilibrium.

Exp Brain Res (2008) 190:153–163 155

123

We obtained a measure of cocontraction as follows. On

each trial and for each antagonistic pair of muscles (for

example, biceps long head and triceps lateral head), the

minimum normalized muscle activity (from the two EMG

signals) was calculated at each point in time (Thoroughman

and Shadmehr 1999; Gribble et al. 2003). This measure of

cocontraction was calculated over the course of movement

to yield a cocontraction trajectory for all trials in the null,

low force and high force conditions.

We also computed a measure of reciprocal activation for

each trial and each pair of antagonist muscles. The measure

of reciprocal activation was the total activity in each

muscle pair minus the activation in each muscle due to

muscle cocontraction. The total activity in each muscle pair

was thus the sum of reciprocal activation plus two times

muscle cocontraction (reflecting the contribution of

cocontraction to the measured activity of each individual

muscle). The contribution following learning of muscle

cocontraction to total muscle activity was calculated as

follows (averaged over the final 25 trials in each experi-

mental condition, not including catch-trials and the

immediately following trial):

C ¼ 2coc

2cocþ recipð3Þ

where C is the proportion of total muscle activity due to

cocontraction, coc is the contribution of each muscle to

cocontraction and recip is the total reciprocal activation as

defined above.

For purposes of statistical analysis, each individual co-

contraction trajectory was divided to four parts, 250 to 50

ms before movement onset, 50 ms before movement onset

to maximum tangential hand velocity, maximum velocity

to end of movement and a 200 ms interval immediately

following movement end. Repeated measures ANOVA and

Bonferroni corrected post hoc tests were used to evaluate

statistical differences.

We also classified muscles according to their role in

movement and cocontraction measures were divided on

this basis. We refer to the cocontraction activity of antag-

onistic muscles that are involved in the generation of

movement as prime mover cocontraction, and the cocon-

traction activity of antagonistic muscles at the largely

stationary joint as stabilizer cocontraction. Previous studies

have demonstrated that activity in biarticular muscles is

closely related to movement and torque at the elbow (Gomi

and Osu 1998; Gribble and Ostry 1998). Therefore, we

have grouped double joint muscles with elbow muscles for

purposes of this analysis. Thus, we grouped together as

stabilizers, the shoulder cocontraction values in the elbow

movement condition and the elbow and biarticular cocon-

traction values in the shoulder movement condition. We

grouped as prime movers, shoulder cocontraction in the

shoulder movement condition and elbow and biarticular

cocontraction in the elbow movement condition.

We tested six additional subjects in a control study that

involved only the shoulder movement condition, and was

conducted without catch trials. Subjects first completed 450

trials under null conditions followed by 450 further trials in

the high force condition. EMG and movement kinematics

were recorded as described above. We verified that per-

formance had reached asymptotic levels by dividing the

experiment into bins of 25 trials each. A repeated measures

ANOVA assessed differences in both cocontraction and

movement curvature over the course of the null and force-

field trials in the control study.

Results

The aim of this study was to assess the role of cocon-

traction following adaptation to a novel dynamic

environment. In order to quantify learning we calculated

the maximum perpendicular deviation (PD) of the hand

from a straight line connecting movement start and end.

Figure 1a shows the mean PD (±SE) for the shoulder

movement condition over the course of the four blocks of

the experiment (50 trials under null-field, 150 trials under

low force-field conditions, 150 trials under high force-

field conditions and 50 final after-effect trials). The

occurrence of catch-trials shown in red was random in the

actual experiment however for visualization purposes they

are plotted at equal intervals. It can be seen that move-

ments are straight during null field trials. When the force-

field is unexpectedly introduced, the hand path deviates

from a straight line but with practice the deviation is

reduced. As in other studies of force-field learning even at

asymptote some curvature remains (Lackner and Dizio

1994; Caithness et al. 2004). The hand path is deviated in

the opposite direction during catch trials and the magni-

tude of the after-effect increases as a function of load.

Following the completion of the low force condition

subjects rest for 5 min and then start the high force

condition in which the direction of force-field remains the

same but the magnitude is increased. As in the low force

condition, the hand initially deviates from a straight line.

Curvature is reduced with practice, however, the effect of

the force-field is not fully removed and some residual

curvature is observed that is greater in magnitude than in

the low force condition. The final block of the experiment

involves 50 trials under null field conditions. Here one

can observe a considerable after-effect that gradually

returns to null field levels. Figure 1b shows the perpen-

dicular deviation for the elbow movement task. The effect

is qualitatively similar to that observed in the shoulder

movement condition.

156 Exp Brain Res (2008) 190:153–163

123

Figure 1 shows that performance reached asymptotic

levels in all conditions. This was examined quantitatively

by assessing changes in perpendicular deviation over the

course of the final 50 trials of training. For this purpose, we

divided the last 50 trials of the low and high force condi-

tions into five bins of ten trials each (catch trials were

excluded). A repeated measures ANOVA found no dif-

ferences in mean curvature over this set of 50 trials (P =

0.30), although, as noted above curvature was reliably

higher in the high force condition (P\0.01). This analysis

thus shows that performance had reached asymptotic levels

long before the end of training.

The larger perpendicular deviation that is observed in

high force-field trials at the end of the learning phase is

accompanied by larger perpendicular deviation in catch-

trials in this condition. Thus, while limb deflections due to

the presence of the force-field are greater, the amount of

learning as assessed by the magnitude of the catch trial

curvature is greater as well. To assess possible differences

in the amount of motor learning with force-field strength

we calculated a learning index, LI, for both the low force

and the high force conditions (Hwang et al. 2003). The

computation was carried out on a per subject basis using

PD measures for final 10% of trials in each condition.

Differences in the LI were tested using a two-way repe-

ated-measures ANOVA. We found that the LI was similar

in low and high force conditions (P = 0.19), averaging 0.65

and 0.61 for the shoulder movement condition and 0.71 and

0.66 for the elbow movement condition (P = 0.24 for

shoulder vs. elbow movement). This suggests that in each

movement direction subjects learned approximately the

same percentage of the force-field regardless of its actual

magnitude.

Figure 2 shows performance under null field conditions

for a representative subject. Mean shoulder and elbow

rotation are shown along with the EMG activity of six

shoulder and elbow muscles. The data are time aligned to

movement start and the highlighted area shows the period

of movement. There is a clear burst of phasic muscle

activity in agonist muscles before the onset of movement

(triceps lateral and long head for elbow movement and

pectoralis for shoulder movement). This phasic burst is

preceded and followed by coactivation of antagonistic

muscles.

To analyze changes in EMG activity that accompany

force field learning, we removed catch trials and the fol-

lowing force-field trial from statistical analyses.

Normalized EMG activity for the final 25 trials in the null,

low force, and high force conditions was used to calculate

the shoulder, elbow and biarticular cocontraction. To

ensure that there were no changes to EMG activity during

these intervals as a result of variation in movement kine-

matics, we examined the maximum tangential hand

velocity during these trials in the three conditions that we

measured. In the shoulder movement condition, maximum

hand velocity was 41 ± 2, 40 ± 2 and 41 ± 3 cm/s for null,

low and high force conditions, respectively. In the elbow

movement condition, maximum hand velocity for the same

three conditions was 43 ± 2, 45 ± 5 and 47 ± 3 cm/s,

respectively. A two-way repeated measures ANOVA

showed no reliable differences in hand velocity for the

different force conditions (P = 0.06). Therefore, changes in

EMG pattern are not present as a byproduct of velocity

change.

Muscle cocontraction was calculated over the course of

movement as the minimum normalized muscle activity for

each antagonistic muscle pair and averaged over the final

25 trials in each condition, that is, at asymptotic perfor-

mance levels following learning. Each resulting

cocontraction trajectory was divided into four parts, 250–

50 ms before the onset of movement, 50 ms before

movement to maximum tangential velocity, maximum

tangential velocity to movement end and the 200 ms

interval following the end of movement. Figure 3 shows

the mean (±SE) normalized cocontraction of shoulder,

elbow and biarticular muscles in each of these four inter-

vals, respectively. Cocontraction for the shoulder

movement condition is shown in green while dark blue

represents cocontraction for the elbow movement

condition.

0 100 200 300 400

Trial Number

0 100 200 300 400

Trial Number

Per

pend

icul

ar D

evia

tion

(mm

)

10

20

30

0

-10

-20

-30

NullField

Low Force

High Force

After Effect

10

20

30

0

-10

-20

Per

pend

icul

ar D

evia

tion

(mm

)

NullField

Low Force

High Force

AfterEffect

A BFig. 1 Maximum perpendicular

deviation (PD) during the four

phases of the experiment. Mean

values across subjects (±SE)

are shown. a Shoulder

movement condition. b Elbow

movement condition. PD

reaches at asymptotic levels

well before the end of training

Exp Brain Res (2008) 190:153–163 157

123

A four-way repeated measure ANOVA followed by

Bonferroni corrected post hoc tests was used to compare

the mean changes in muscle cocontraction. For this par-

ticular analysis we divided muscles based on their role in

the movement. We will refer to muscles at the largely

stationary joint as stabilizers and muscles involved in

generating limb displacement as prime movers. The anal-

ysis showed that cocontraction increased with magnitude

of the force-field (P\0.001). Cocontraction magnitudes at

each force level were found to be reliably different from

one another by post hoc tests (P \ 0.02 in all cases).

Cocontraction magnitudes also varied over the course of

Shoulder Movement

Elbow Movement

Deltoid

Pectoralis

Triceps Lateral

Biceps Long

Triceps long

Biceps Short

20 Deg

0.2 sec

20 µV

Shoulder Movement

Elbow Movement

Deltoid

Pectoralis

Triceps Lateral

Biceps Long

Triceps long

Biceps Short

A BFig. 2 Shoulder and elbow

rotation and mean EMG activity

of six shoulder and elbow

muscles for a representative

subject. a Shoulder movement

condition. b Elbow movement

condition

Mea

n N

orm

aliz

ed C

ocon

trac

tion(

%)

Shoulder

60

40

20

Null FF1 FF2

30

60

90

Null FF1 FF2

Shoulder

60

100

140

Null FF1 FF2

Shoulder

40

80

120

Null FF1 FF2

Shoulder

60

Elbow

40

20

Null FF1 FF2

30

60

90

Null FF1 FF2

Elbow

60

100

140

Null FF1 FF2

Elbow

40

80

120

Null FF1 FF2

Elbow

Biarticular

60

40

20

Null FF1 FF2

30

60

90

Null FF1 FF2

Biarticular

60

100

140

Null FF1 FF2

Biarticular

40

80

120

Null FF1 FF2

Biarticular

Force-Field Conditions

A B C DFig. 3 During wholly stable

interactions with the

environment, cocontraction is

modulated with the level of the

force-field and also with the

direction of movement. Mean

across subjects (±SE) of

shoulder, elbow, and biarticular

cocontraction for shoulder

movement direction (green) and

elbow movement direction

(dark blue). Note that for

visualization purposes different

scales have been used. a From

250 to 50 ms before movement.

b A measure of 50 ms before

movement to maximum

velocity. c Maximum velocity

to movement end. d A measure

of 200 ms following movement

end

158 Exp Brain Res (2008) 190:153–163

123

the movement (P \ 0.001). Cocontraction was lowest in

the intervals preceding peak velocity and increased reliably

to a maximum in the deceleration phase of movement (P\0.05 in both movement directions). The cocontraction level

at the end of movement was not reliably different from that

in the deceleration phase (P = 0.08).

The relative magnitude of cocontraction in stabilization

versus movement related muscles also varied during

movement (P \ 0.01). In the period prior to movement

(Fig. 3a) and in the interval from movement start to peak

velocity (Fig. 3b), post hoc comparisons showed that

cocontraction magnitudes were reliably greater for muscles

involved in joint stabilization than for muscles involved

primarily in generating the movement (P\0.01 and\0.05,

respectively). In the two subsequent intervals, from peak

velocity onward, cocontraction levels were similar for

stabilization muscles and prime movers (P = 0.24, 0.07

respectively).

Muscle cocontraction accounted for a substantial portion

of total activity in each muscle pair even following learn-

ing. Figure 4a shows a representative example of muscle

activity in the shoulder movement condition (null field).

The records display an interval from 500 ms before

movement start to 500 ms after movement end. The

activity for each muscle pair is shown separately. Flexor

muscles are shown in green, extensors are in blue and the

common portion due to cocontraction is the light blue

shaded region.

We quantified the proportion of total muscle activity due

to cocontraction, C, for each of the four time intervals

shown in Fig. 3. Figure 4b shows this ratio in each of the

four movement phases and the three force-field conditions

(averaged over muscle pairs). Overall it can be seen that

cocontraction accounted for approximately half of the

measured muscle activity, except in the acceleration phase

of movement where the contribution is less. It can also

been seen the proportion of activity due to cocontraction is

relatively constant in the different force conditions tested

here and in different movement directions. Thus while both

phasic muscle activity and cocontraction increase with

level of the force-field the overall proportion is for the most

part unchanged.

Differences in C were assessed using a three-way

repeated measures ANOVA followed by Bonferroni cor-

rected post hoc tests. We found a significant change in the

proportion of muscle activation due to cocontraction over

the four time intervals of the movement (P \ 0.001).

Cocontraction during the acceleration phase of movement

was reliably less than in the three other intervals (P\0.001

in each case). Overall, cocontraction was found to account

for 53, 36, 57 and 56% of the total activity in these inter-

vals, respectively. The lowest value, in the acceleration

phase of the movement, is due to the large phasic muscle

activation associated with initiation of limb movement. We

found that the proportion of cocontraction did not differ

with force level (P = 0.06). However there were reliable

0.2 sec

Shoulder

Elbow

Biarticular

Interval 1

Interval 2Interval 3interval 4

Force-Field Level

0

0.2

0.4

0.6

0.8

0

0.2

0.4

0.6

0.8

Null FF1 FF2

Null FF1 FF2

C R

atio

Elbow Condition

Shoulder Condition

A B

Fig. 4 Cocontraction accounts for a substantial proportion of total

muscle activity. a Representative example of muscle activity in the

shoulder movement condition for an individual subject (averaged

over 20 null field trials). Agonist muscles are shown in green,

antagonist muscles are in dark blue. The darker shaded area shows

the defined cocontraction activity for each muscle pair. The verticalhighlighted band indicates the period of movement. b Mean across

subjects (±SE) of the relative contribution of cocontraction to total

muscle activity (C ratio), averaged over muscles. The figure shows

the four movement intervals at each of the three force-field levels

Exp Brain Res (2008) 190:153–163 159

123

differences in the proportion of cocontraction with move-

ment direction (P \ 0.01). A greater proportion of

cocontraction overall was observed in the elbow movement

condition but the differences were small (52.5 vs. 48.5%).

There were two other statistically reliable differences.

Cocontraction during the acceleration phase of movement

accounted for less of the total in the shoulder condition

(32.5 vs. 40.5%). Similarly the proportion of cocontraction

at the end of movement was less in the shoulder condition

(53.7 vs. 59.3%) (P \ 0.02 in each case).

Figure 5 shows changes in cocontraction over the course

of learning. The mean (±SE) cocontraction level is given

during the interval from 250 to 50 ms before the onset of

movement (the other parts of the cocontraction trajectory

show a similar pattern). Null field trials are shown in dark

blue, cocontraction in the low force condition is shown in

green (with catch trials and the first trial following the

catch trial removed) and cocontraction in the high force

condition is shown in light blue. A repeated measures

ANOVA revealed a reliable difference between the mean

cocontraction activity of the first 10% of trials (the very

first trial was excluded) and the last 10% of trials (P \0.05) in each condition. However, there is still substantial

cocontraction even in the plateau phase of learning.

We verified that EMG activity was at asymptotic levels

by assessing changes in the cocontraction level over the

last 50 trials of force-field learning. As in our earlier

assessment of movement curvature, we divided cocon-

traction measures over the last 50 force-field trials in each

of the high and low force conditions into five bins of ten

trials each (catch trials and the following force-field trial

are excluded). Over the course of these trials we found no

statistically reliable differences in cocontraction activity (P

[ 0.4, repeated measures ANOVA). Cocontraction had

thus reached asymptotic levels long before the end of

training. In summary, the present study finds persistent co-

contraction throughout all phases before and after learning

a novel task.

20

60

100

Shoulder

20

60

100Elbow

50 100 150 200

50 100 150 200 50 100 150 200

50 100 150 20050 100 150 200

20

60

100

Biarticular

50 100 150 200

20

60

100

20

60

100

20

60

100

Mea

n N

orm

aliz

ed C

ocon

trac

tion

Shoulder

Elbow

Biarticular

Trial Number

A BFig. 5 Cocontraction declines

with learning but there is

persistent cocontraction even at

asymptotic performance levels.

Mean across subjects (±SE) of

shoulder, elbow, and biarticular

muscle cocontraction from 250

to 50 ms before movement. Null

is in blue, low force is in greenand high force is in light blue(catch trials and the first force-

field trial afterwards are

removed). a Shoulder condition.

b Elbow condition

160 Exp Brain Res (2008) 190:153–163

123

We conducted a control study with six new subjects to

rule out the possibility that either insufficient learning or

the presence of catch-trials were the source of the persistent

cocontraction. The new subjects were tested over the

course of 450 null movements and 450 high force condition

movements, without any catch-trials. Figure 6a shows

normalized cocontraction, over the interval 250 ms before

movement start to 200 ms after movement end, averaged

over muscle pairs and subjects. Cocontraction is seen to be

at asymptotic levels at the end of each phase of the

experiment and greater in magnitude in the high force

condition. We examined this quantitatively by dividing the

cocontraction measures in both null and force field condi-

tions into five bins of 25 trials each that were spaced evenly

throughout the 450 trial interval. Differences were assessed

using a two-way repeated measures ANOVA. We found

that cocontraction was reliably greater under high force

conditions (P \ 0.001). Using Bonferroni corrected pair-

wise comparisons, we found no differences in

cocontraction over the final four bins, that is, following the

first 100 training trials (P [ 0.1 in all cases). This shows

that even after a longer period of training without catch

trials subjects consistently use cocontraction as part of the

control that underlies movement.

Figure 6b shows that the proportion of total muscle

activity due to cocontraction remains high regardless of the

force-field condition. Indeed, even in the absence of catch

trails, cocontraction accounts for the same proportion of

total muscle activity (in range of 50%). There were no

reliable differences over the final 25 trials in the null and

force-field conditions in the proportion of activity due to

cocontraction (P = 0.09). The mean (±SE) of C ratio in the

null condition was 0.45 ± 0.03 while the same value for the

force-field condition was 0.55 ± 0.05. Figure 6c shows that

learning as assessed by maximum perpendicular deviation

reaches null field levels within about 100 trials. A repeated

measures ANOVA, divided as above into bins of 25 trials,

indicated no differences in movement curvature over the

final 350 training movements (P [ 0.1).

The order in which subjects performed the task requires

comment. In all cases, null field trials were followed by

low force trials and then by high force trials. Although

subjects were given 5–10 min rest breaks between the three

conditions, the possibility remains that muscle fatigue

played a role in the observed results. We tested for muscle

fatigue quantitatively. It is known that during isometric

contraction the median of the EMG power spectrum shifts

toward lower frequencies. Recently time–frequency anal-

ysis has been used to assess muscle fatigue in cyclic

dynamic contractions (Bonato et al. 2001). It has been

shown that as a muscle fatigues the Instantaneous median

frequency (IMDF) shifts toward lower values. To test for

fatigue in the present study we compared the mean IMDF

of the last ten trials in each condition. A three-way repeated

NullField

HighForce

NullField

HighForce

0 200 400 600 8000

50

100

150

200

250

0 200 400 600 8000.3

0.4

0.5

0.6

0 200 400 600 80010

0

10

20

30

40

Trial Number

Trial Number

Trial Number

Per

pend

icul

ar D

evia

tion

(mm

)M

ean

Nor

mal

ized

Coc

ontr

actio

n(%

)

C R

atio

NullField

HighForce

B

C

AFig. 6 Cocontraction remains

high after extensive training in

the absence of catch-trials.

a Mean normalized

cocontraction across subjects

and muscle pairs (±SE) over the

interval 250 ms before

movement start to 200 ms after

movement end. The level of

cocontraction is greater in the

high force condition. b Mean Cratio (proportion of total muscle

activity due to cocontraction)

averaged over subjects and

muscle pairs (±SE). In both null

and high force conditions,

cocontraction accounts for

about 50% of total muscle

activity. c Mean perpendicular

deviation (±SE) over the course

of training. Performance in the

high force condition approaches

null field levels within about

100 trials

Exp Brain Res (2008) 190:153–163 161

123

measures ANOVA (two directions of movement, three

different force-field conditions, six muscles) found no

differences in the mean IMDF for the three force-field

conditions (P = 0.16). Muscle fatigue can thus be ruled out

as a potential source of the present results.

Discussion

We have assessed characteristics of antagonistic muscle

cocontraction over the course of dynamics learning. Sub-

jects were given extensive practice in producing reaching

movements to two visual targets. Each subject was tested

in a null field followed by two force-field conditions, a low

force and then a high force condition. Muscle coactivation,

as assessed by the overlap of agonist and antagonist muscle

activity was observed in all phases of movement and

throughout the course of training and accounted for a

substantial proportion of total muscle activation even fol-

lowing learning. Importantly, cocontraction characteristics

varied with the specific details of the task. In particular,

cocontraction varied in magnitude with the strength of the

force-field and also varied with movement direction. In

each direction, cocontraction changed over the course of

movement such that before movement start and up to peak

velocity, cocontraction in muscles involved primarily in

joint stabilization was higher than in the prime movers.

Following peak velocity cocontraction activity patterns in

stabilizer muscles and prime movers was similar.

It has been suggested that a combination of feedforward

and impedance control are involved in adaptation to both

stable and unstable dynamical environments (Franklin et al.

2003b). A number of studies have evaluated patterns of

muscle activity in the early stages of learning (Franklin

et al. 2003b; Thoroughman and Shadmehr 1999). These

studies report an initial increase in activity for all muscles

followed by a decrease in cocontraction as learning pro-

gresses. The data from the present study (Fig. 5) show a

similar decline. The present study extends the previous

findings by showing that the cocontraction which remains

following motor learning is tuned to the requirements of

the task such that it varies in magnitude in a systematic

fashion both with force level and movement direction.

Moreover, we observe that the relative contribution of

cocontraction to total muscle activity remains essentially

the same following learning at all force levels. This shows

that cocontraction control is involved in all phases of

learning and that the nervous system maintains a fairly

constant balance of cocontraction and reciprocal activation.

The relative level of cocontraction is more or less con-

stant over different phases of movement as well. Except for

the acceleration phase of the movement (from just before

movement start to maximum velocity), we found that

cocontraction quite consistently accounted for 50–60% of

total muscle activity. The reason that cocontraction per-

centages were not higher in the intervals just before and

just after movement is the presence of phasic muscle

activity both prior to movement onset and persistent

activity following movement end (Suzuki et al. 2001). In

the present dataset, the onset of phasic activity is clearly

evident by 200 ms before the start of movement and is

likewise present for several hundred milliseconds after the

end of the movement (see Fig. 4a). This accounts for the

fact that the observed proportion of cocontraction in these

intervals (250–50 ms before movement start and in the 200

ms following movement end) is comparable to that

observed during movement.

The directional tuning of muscle cocontraction and

corresponding directional changes to hand stiffness have

been reported previously (Burdet et al. 2001; Darainy et al.

2004; Franklin et al. 2003b; Gomi and Osu 1998; Perreault

et al. 2002). However, the studies in which this directional

tuning has been observed, have involved either adaptation

to an unstable dynamic environment or biofeedback of

muscle cocontraction that is presented to the subject. The

present results (Fig. 3, panels a, b) show that different

patterns of shoulder and elbow cocontraction, and hence

presumably, different underlying neural commands, are

involved for different movement directions even when the

interaction of the hand with the environment is stable.

Cocontraction magnitudes were greater at the stationary

joint. This difference presumably arises as a consequence

of the biomechanics of the experimental design. The force-

field here produces torques that act primarily about the

stationary joint. The observed difference in cocontraction

between the stationary and moving joint may well reflect

the need for greater stability under these conditions.

The way in which muscle cocontraction has been

assessed in this study merits comment. The rationale for

using overlap in EMG levels of antagonistic muscles as a

measure of cocontraction is that the activity that is com-

mon to both muscles results in changes in impedance

without producing accompanying changes in net joint tor-

que. However the measurement of cocontraction in this

manner also has certain limitations. First cocontraction as

assessed in this way may in part reflect phasic muscle

activation. This component is of course absent when co-

contraction is assessed under stationary conditions.

Nevertheless, even if the measured cocontraction in the

present study derives in part from phasic activation, it acts

to increase impedance and in this sense its effects are

similar to those observed for coactivation under static

conditions. A second caveat is that cocontraction in the

present study is estimated over time by computing the

minimum normalized EMG activity of antagonistic muscle

pairs. For each muscle, EMG activity is normalized relative

162 Exp Brain Res (2008) 190:153–163

123

to the observed EMG level under static conditions well

before the movement onset. However, the normalization

that serves to cancel out each opposite muscle torque under

static conditions may require modification during move-

ment. For example, changes in muscle moment arms along

with length and velocity dependent differences in muscle

force generating abilities (and the activity of other syner-

gistic muscles) may influence torque and the actual level of

muscle cocontraction. It is worth noting that measures

taken in the interval preceding movement onset (Fig. 3a)

are not contaminated by changes in muscle length and

velocity. Similarly, measures from muscles that are pri-

marily involved in joint stabilization are less affected.

We also wish to comment on a potential concern

regarding the muscle activity level that we have used for

EMG normalization. If EMG values in the interval used for

normalization were small relative to those observed during

movement, the effects of measurement error or noise could

be greatly exaggerated. In the present study, EMG signals

were normalized relative to the mean muscle activity in the

high force condition during a 200 ms window before

movement onset. The mean normalization factor (over

subjects and muscles) was 11 lV whereas the mean base-

line muscle activity was 1.7 lV. Muscle activity in the

interval used for normalization was thus almost 6.5 times

the baseline activity level. By way of comparison, maxi-

mum EMG activity for the null field movements shown for

a representative subject in Fig. 2 was around 40 lV. Our

normalization factor is thus large in relation to background

signal levels and well within the range of signals that is

applied to for normalization. It is thus unlikely that this

procedure increases the effects of measurement error or

noise.

Acknowledgments The authors thank Eric Perreault for comments

and Guillaume Houle and Andrew Mattar for technical assistance.

This research was supported by NICHD Grant HD-48924, NSERC

Canada, and FQRNT Quebec.

References

Bonato P, Roy SH, Knaflitz M, De Luca CJ (2001) Time–frequency

parameters of the surface myoelectric signal for assessing

muscle fatigue during cyclic dynamic contractions. IEEE Trans

Biomed Eng 48:745–753

Burdet E, Osu R, Franklin DW, Milner TE, Kawato M (2001) The

central nervous system stabilizes unstable dynamics by learning

optimal impedance. Nature 414:446–449

Caithness G, Osu R, Bays P, Chase H, Klassen J, Kawato M, Wolpert

DM, Flanagan JR (2004) Failure to consolidate the consolidation

theory of learning for sensorimotor adaptation tasks. J Neurosci

24:8662–8671

Darainy M, Malfait N, Gribble PL, Towhidkhoh F, Ostry DJ (2004)

Learning to control arm stiffness under static conditions.

J Neurophysiol 92:3344–3350

Franklin DW, Burdet E, Osu R, Kawato M, Milner TE (2003a)

Functional significance of stiffness in adaptation of multijoint

arm movements to stable and unstable dynamics. Exp Brain Res

151:145–157

Franklin DW, Osu R, Burdet E, Kawato M, Milner TE (2003b)

Adaptation to stable and unstable dynamics achieved by

combined impedance control and inverse dynamics model.

J Neurophysiol 90:3270–3282

Gomi H, Osu R (1998) Task-dependent viscoelasticity of human

multijoint arm and its spatial characteristics f or interaction with

environments. J Neurosci 18:8965–8978

Gribble PL, Ostry DJ (1998) Independent coactivation of shoulder

and elbow muscles. Exp Brain Res 123:355–360

Gribble PL, Mullin LI, Cothros N, Mattar A (2003) Role of

cocontraction in arm movement accuracy. J Neurophysiol

89:2396–2405

Hwang EJ, Donchin O, Smith MA, Shadmehr R (2003) A gain-field

encoding of limb position and velocity in the internal model of

arm dynamics. PLoS Biol 1:E25

Lackner JR, DiZio P (1994) Rapid adaptation to Coriolis force

perturbations of arm trajectory. J Neurophysiol 72:299–313

Milner TE, Cloutier C (1993) Compensation for mechanically

unstable loading in voluntary wrist movement. Exp Brain Res

94:522–532

Perreault EJ, Kirsch RF, Crago PE (2002) Voluntary control of static

endpoint stiffness during force regulation tasks. J Neurophysiol

87:2808–2816

Shadmehr R, Mussa-Ivaldi FA (1994) Adaptive representation of

dynamics during learning of a motor task. J Neurosci 14:3208–

3224

Suzuki M, Shiller DM, Gribble PL, Ostry DJ (2001) Relationship

between cocontraction, movement kinematics and phasic muscle

activity in single joint arm movement. Exp Brain Res 140:171–

181

Thoroughman KA, Shadmehr R (1999) Electromyographic correlates

of learning an internal model of reaching movements. J Neurosci

19:8574–8588

Exp Brain Res (2008) 190:153–163 163

123